JP7602477B2 - Oil deterioration prediction device, oil deterioration prediction system, oil deterioration prediction method, oil exchange system, and fryer system - Google Patents

Description

The present invention relates to a deterioration prediction device, a deterioration prediction system, a deterioration prediction method, an oil exchange system, and a fryer system for predicting the degree of deterioration of oils and fats.

The edible oils and fats used in cooking fried foods deteriorate when cooking food repeatedly, and so need to be replaced at appropriate times. In order to objectively determine when to replace such oils and fats, devices that detect and judge the color, viscosity, odor, etc. of the oils and fats are known.

For example, the detection device in the following Patent Document 1 has a sensor unit attached to a ventilation fan above an oil tank. This sensor unit has a sensitive membrane that adsorbs gas molecules that are the source of odors and a transducer that converts the gas molecules attached to the sensitive membrane into an electrical signal, and detects odors emanating from the cooking oil. The control unit of the detection device then determines the degree of deterioration of the cooking oil based on information about the odor detected by the sensor unit during deep-frying and the type of food cooked with the cooking oil (paragraphs 0017, 0021, Figure 1).

Patent No. 6448811

However, in the case of the detection device of Patent Document 1, various odors exist in the kitchen other than the ingredients being cooked (such as fragrant odors and burnt odors emanating from foods other than fried foods), making it difficult to accurately predict the deterioration of cooking oil from odor alone.

The present invention has been made in consideration of the above circumstances, and has an object to provide a deterioration prediction device that can predict deterioration of oils and fats simply and accurately.

The deterioration prediction device of the present invention is a deterioration prediction device that predicts the degree of deterioration of edible oils and fats, and is characterized in that it includes an acoustic data acquisition unit that acquires acoustic data when fried foods are cooked using the oil and fat contained in an oil vat, an index extraction unit that extracts indexes related to the deterioration of the oil and fat from the acoustic data acquired by the acoustic data acquisition unit, and a judgment unit that judges the degree of deterioration of the oil and fats based on the indexes extracted by the index extraction unit.

The acoustic data acquisition unit of the deterioration prediction device acquires acoustic data of oils and fats when cooking deep-fried foods such as tempura. The index extraction unit extracts various acoustic components, such as the frequency average and the frequency standard deviation, from the acoustic data as indexes related to deterioration of the oil and fats. The determination unit then determines the degree of deterioration of the oil and fats, i.e., whether deterioration has progressed with use, based on the indexes. This allows the device to easily and accurately predict deterioration of oil and fats.

In the deterioration prediction device of the present invention, it is preferable that the deterioration prediction device further includes an alarm unit that notifies the degree of deterioration of the oil or the timing for replacing the oil, and that the alarm unit makes the alarm when the judgment unit judges that a predetermined replacement threshold has been exceeded based on the degree of deterioration of the oil.

According to this configuration, the notification unit of the deterioration prediction device notifies the user of the deterioration level of the oil, so that the user can grasp the usage status of the oil. In addition, the notification unit notifies the user of the timing to change the oil from a predetermined threshold based on the judgment result of the judgment unit, so that the user can change the oil at an appropriate timing. Here, the "timing to change the oil" may be the actual timing to change the oil, or the remaining usable time estimated from the current deterioration level of the oil.

In addition, in the deterioration prediction device of the present invention, it is preferable that the index is one or more selected from frequency mean, frequency standard deviation, frequency median, frequency standard error, frequency mode, frequency first quartile, frequency third quartile, frequency interquartile range, frequency centroid, frequency skewness, frequency kurtosis, frequency spectrum flatness, frequency spectrum entropy, frequency spectrum precision, acoustic complexity index, acoustic entropy, and dominant frequency.

According to this configuration, one or more of the indices that are highly correlated with deterioration of the oil or grease are selected, thereby enabling the present device to predict the deterioration with high accuracy.

The deterioration prediction system of the present invention is a deterioration prediction system that predicts the degree of deterioration of edible oils and fats, and comprises a detection device and a machine learning device. The detection device comprises an acoustic data acquisition unit that acquires acoustic data when cooking fried foods using the oil contained in an oil tank, a memory unit that stores a learning model created by the machine learning device that can determine the deterioration of the oil and fats, and a judgment unit that uses the learning model to determine the degree of deterioration of the oil and fats. The machine learning device comprises a learning model creation unit that extracts indicators related to the deterioration of the oil and fats from the acoustic data acquired by the acoustic data acquisition unit, performs machine learning by linear regression using the indicators, and creates the learning model.

The deterioration prediction system of the present invention is composed of a detection device and a machine learning device. In the detection device, an acoustic data acquisition unit acquires acoustic data of oil and fat when cooking fried foods, and a determination unit determines the degree of deterioration of the oil and fat using a learning model.

The learning model creation unit then extracts indicators related to the deterioration of oils and fats from the acquired acoustic data and performs machine learning using linear regression. This updates the learning model, enabling the system to easily and accurately predict the deterioration of oils and fats.

In the deterioration prediction system of the present invention, the linear regression is preferably one or more types selected from simple regression, multiple regression, partial least squares (PLS) regression, and orthogonal projection partial least squares (OPLS) regression.

To create the learning model, linear regression such as simple regression, multiple regression, partial least squares (PLS) regression, orthogonal projection partial least squares (OPLS) regression, etc., is used. This allows the system to create a learning model that can accurately determine the deterioration of oils and fats.

In the deterioration prediction system of the present invention, it is preferable that the detection device and the machine learning device are integrated together.

For example, by installing the integrated deterioration prediction system of the present invention near an oil tank in a store or factory, a user can obtain prediction results of deterioration of oils and fats on the spot.

In addition, in the deterioration prediction system of the present invention, it is preferable that the detection device is installed near the oil tank in a store or factory, and the machine learning device is installed in a remote location away from the store or factory.

The detection device having an acoustic data acquisition unit and the like is installed near an oil tank in a store or factory, but the machine learning device may be installed in a remote location of the store. Since the machine learning device is a device separate from the detection device, the learning model created by the machine learning device can be acquired by communication or the like.

In addition, in the deterioration prediction system of the present invention, it is preferable that the detection device includes a first communication unit that transmits the acoustic data acquired by the acoustic data acquisition unit to the machine learning device, and the machine learning device includes a second communication unit that receives the acoustic data from the detection device.

The detection device includes a first communication unit and transmits acoustic data to the machine learning device. The machine learning device includes a second communication unit and receives the acoustic data and performs machine learning. When the detection device and the machine learning device are separate devices, the system can share the necessary work by transmitting and receiving data via each communication unit.

In the deterioration prediction system of the present invention, it is preferable that the first communication unit and the second communication unit are capable of wireless communication.

Since the detection device can transmit acoustic data to the machine learning device via wireless communication via the first communication unit, the functions of the detection device can be minimized to make the detection device smaller.

The deterioration prediction method of the present invention is a deterioration prediction method for predicting the degree of deterioration of edible oils and fats, and is characterized by comprising an acoustic data acquisition step for acquiring acoustic data when cooking fried foods using the oil and fats, an index extraction step for extracting indexes related to the deterioration of the oil and fats from the acoustic data acquired in the acoustic data acquisition step, and a determination step for determining the degree of deterioration of the oil and fats based on the indexes extracted in the index extraction step.

In the deterioration prediction method of the present invention, in the acoustic data acquisition step, acoustic data of oils and fats when cooking deep-fried foods such as tempura is acquired. Then, in the index extraction step, various acoustic components such as frequency average and frequency standard deviation are extracted from the acoustic data as indexes related to deterioration of the oils and fats. Furthermore, in the determination step, the degree of deterioration of the oils and fats, i.e., whether deterioration has progressed with use, is determined based on the index. This allows the present method to easily and accurately predict deterioration of oils and fats.

The oil and fat exchange system of the present invention is characterized in that, based on the notification information regarding the degree of deterioration of the oil and fat output from the above-mentioned deterioration prediction device, it performs one or more of the following: a) notifying an oil and fat distributor and ordering new oil and fat; b) notifying an oil and fat manufacturer and drawing up a production plan or sales plan for the oil and fat; c) notifying the store or factory headquarters or the oil and fat manufacturer and proposing or instructing the store or factory it supervises on how to use the oil and fat; d) notifying a waste oil collection company or oil and fat manufacturer and arranging for collection of waste oil; and e) notifying a cleaning company and arranging for cleaning of the oil tank.

In the oil and fat exchange system of the present invention, based on the notification information on the deterioration degree of the oil and fat, for example, when the notification information is notified a predetermined number of times, a notification is sent to an oil and fat distributor to order new oil and fat. Also, based on the notification information, a notification is sent to an oil and fat manufacturer to create a production plan or sales plan for the oil and fat. In this way, the system can establish a production and sales plan according to the oil and fat exchange pace.

In addition, the oil exchange system notifies the headquarters of the store or factory or the oil manufacturer based on the notification information, and proposes or instructs the stores or factories under its control on how to use the oils and fats. For example, the headquarters instructs each store to use oils and fats efficiently and by replacing them as needed. Furthermore, based on the notification information, the system notifies a waste oil collector to arrange for the collection of the waste oil, and notifies a cleaning company to arrange for the cleaning of the oil tank, so that the system can quickly carry out the process from supplying the oils and fats to disposing of the oil.

The fryer system of the present invention is characterized in that it comprises a valve control unit that controls a valve provided in an oil tank based on notification information regarding the degree of deterioration of the oil output from the above-mentioned deterioration prediction device, and the valve control unit automatically discards the oil contained in the oil tank.

In the fryer system of the present invention, the valve control unit controls the valve of the oil tank based on the notification information regarding the degree of deterioration of the oil, so that the system can automatically discard the oil during use.

In the fryer system of the present invention, it is preferable that the valve control unit automatically supplies new oil to the oil tank.

According to this configuration, the valve control unit controls the valve to automatically supply new oil to the oil tank, thereby reducing the burden on the user of the series of tasks from checking the degree of deterioration of the oil to supplying used oil and new oil.

According to the present invention, deterioration of fats and oils can be predicted simply and accurately.

1 is a diagram illustrating an overview of a deterioration prediction device and a flyer according to a first embodiment. FIG. 1 is a functional block diagram of a deterioration prediction device according to a first embodiment; 4 is a flowchart showing a process for determining deterioration of frying oil by the deterioration prediction device. FIG. 11 is a functional block diagram of a deterioration prediction device (deterioration prediction system) according to a second embodiment. FIG. 13 is a graph showing the relationship between a calibration curve obtained by machine learning (simple regression) and the heating time of test data. A table showing the mean and standard deviation of the predicted values in FIG. 5A. FIG. 13 is a graph showing the relationship between a calibration curve obtained by machine learning (multiple regression) and the acid value of test data. A list of the actual measured values, average predicted values, and standard deviations for FIG. 6A. FIG. 13 is a graph showing the relationship between a calibration curve obtained by machine learning (OPLS) and the heating time of test data. A table showing the mean and standard deviation of the predicted values in FIG. 7A. FIG. 13 is a graph showing the relationship between a calibration curve obtained by machine learning (OPLS) and the acid value of test data. A list of the actual measured values, average predicted values, and standard deviations for FIG. 8A. FIG. 13 is a diagram showing the relationship between a calibration curve obtained by machine learning (PLS) and the color of test data. A list of the actual measured values, average predicted values, and standard deviations of FIG. 9A. FIG. 13 is a graph showing the relationship between a calibration curve obtained by machine learning (PLS) and the viscosity increase rate of test data. A list of the actual measured values, average predicted values, and standard deviations of FIG. 10A. FIG. 13 is a diagram illustrating an oil and grease exchange system according to a third embodiment. FIG. 13 is a diagram illustrating a fryer system according to a fourth embodiment.

Hereinafter, an embodiment of a deterioration prediction device according to the present invention will be described with reference to the drawings.

[First embodiment]
First, an overview of a deterioration prediction device 1 and a fryer 20 according to a first embodiment of the present invention will be described with reference to Fig. 1. As shown in the figure, the deterioration prediction device 1 is mainly composed of an acoustic data acquisition unit 2 (the "acoustic data acquisition unit" of the present invention) and a processing unit 3. The acoustic data acquisition unit 2 is, for example, a highly directional microphone, and acquires sounds (such as the popping sounds of bubbles) made when cooking fried foods with frying oil (the "oil and fat" of the present invention) contained in the fryer 20.

The acquired sound (hereinafter, sound data) is transmitted to the processing unit 3. Then, the processing unit 3 extracts features and analyzes the deterioration of the frying oil from the features. The processing unit 3 includes a display unit 5, a control unit 10, etc., which will be described in detail later.

The fryer 20 has a box-shaped cabinet 21 and is provided with an oil tank 22 for storing frying oil therein. The temperature of the frying oil stored in the oil tank 22 can be adjusted by a heater 23. For example, when cooking croquettes, the frying oil is adjusted to 180°C.

An oil drain pipe 25 is connected to the bottom of the oil tank 22 via a valve 24. The bottom of the oil tank 22 is shaped like a funnel, inclined downward to facilitate drainage of the oil. Deteriorated frying oil is discharged as waste oil by opening the valve 24. A waste oil tank 26 is disposed below the oil drain pipe 25 to store the waste oil.

The oil vat 22 is assumed to be used for a large fryer used in a restaurant, a bar, etc., but is not limited to this. That is, the oil vat 22 may be used for a smaller fryer or may be a home frying cooker.

In this embodiment, the acoustic data acquisition unit 2 is installed at a height of about 1 m from the fryer 20 (diagonally above the oil vat 22). Since oil smoke is usually generated during cooking, a ventilator is installed above the fryer 20 to exhaust the oil smoke outside (not shown). The acoustic data acquisition unit 2 may be attached to the side of the ventilator. The acoustic data acquisition unit 2 may be installed near the oil vat 22 on the side, wall, ceiling, or the like of the cabinet 21.

FIG. 2 is a functional block diagram of the deterioration prediction device 1 according to the first embodiment.

The deterioration prediction device 1 is composed of an acoustic data acquisition unit 2 and a processing unit 3, and the processing unit 3 has an input unit 4, a display unit 5, a storage unit 6, a notification unit 7, and a control unit 10. First, the acoustic data acquisition unit 2 acquires sounds generated when cooking croquettes, tempura, etc.

The sound data acquisition unit 2 may use a microphone, or may record sound using a video camera or a recording function of a smartphone. For example, the sound data acquisition unit 2 acquires sound data of the cooking time of the deep-fried ingredients at an audio sample rate of 48 kHz. In addition, since the operation sounds caused by putting in and taking out the deep-fried ingredients become noise in the sound data, the sounds for 10 seconds after the start of recording and 10 seconds before the end are cut.

When frying oil deteriorates, fatty acids contained in the frying oil are decomposed, and the sound generated during cooking gradually changes. The index extraction unit 11 of the control unit 10 extracts an index related to deterioration of frying oil (hereinafter, index data) from the acquired sound data, and the result acceptance unit 12 accepts the index data.

As characteristics often appear in the frequency of sounds emitted during cooking, the index data used were the frequency mean (f_mean), frequency standard deviation (f_sd), frequency median (f_median), frequency standard error (f_sem), and frequency mode (f_mode).

Other index data include the first quartile of frequency (f_Q25) at 25% from the minimum frequency, the third quartile of frequency (f_Q75) at 75% from the minimum frequency, the interquartile range of frequency (f_IQR), the center of frequency (f_cent), frequency skewness (f_skewness), frequency kurtosis (f_kurtosis), frequency spectrum flatness (f_sfm), frequency spectrum entropy (f_sh), frequency spectrum precision (prec), acoustic complexity index (d.ACI), acoustic entropy (d.H), and dominant frequency (dfnum). For the acoustic analysis, seewave (Sound Analysis and Synthesis) and ropls (PCA, PLS(-DA) and OPLS(-DA) for multivariate analysis) were used.

Specifically, the control unit 10 is a processor that controls and manages the entire deterioration prediction device 1, and is configured with a CPU (Central Processing Unit) that executes a program that defines a control procedure. Such a program is stored in, for example, the storage unit 6 or another external storage device.

The control unit 10 controls the entire processing unit 3 to execute each process of the deterioration prediction device 1. For example, the control unit 10 starts up the deterioration prediction device 1 based on a predetermined input operation by a user (store clerk). The predetermined input operation is, for example, an operation of turning on the power of the deterioration prediction device 1, and an operation of setting the cooking time and the temperature of the frying oil.

The input unit 4 is various switches that accept input operations from the user, and is composed of, for example, operation buttons, operation keys, etc. The input unit 4 is not limited to this, and may be composed of a touch panel. Furthermore, the input unit 4 accepts a predetermined input operation from the user before the execution of processing by the deterioration prediction device 1, and transmits a signal based on the user's input operation to the control unit 10.

The display unit 5 displays various items for the user to perform input operations. For example, when the user selects the type of food to cook, the display unit 5 displays the type of food based on data related to the type of food stored in the memory unit 6. In addition, when the notification unit 7 notifies the user of the degree of deterioration of the frying oil, the display unit 5 displays a message indicating that replacement is necessary as a secondary notification.

The storage unit 6 is composed of a semiconductor memory, a magnetic memory, or the like, and stores various information and programs for operating the deterioration prediction device 1. The storage unit 6 stores the acquired acoustic data, the learning model, and data related to the food to be cooked. For example, the storage unit 6 stores correlation data indicating the correlation between the acoustic data and the degree of deterioration of the frying oil for each type of food handled. The storage unit 6 also stores threshold information for notification that differs for each type of food.

When it is determined that the deterioration level of the frying oil exceeds a predetermined threshold, the notification unit 7 notifies the user of this fact. In this way, the notification unit 7 notifies the user of the timing to change the frying oil. Here, the "timing to change" refers to the timing to actually change the frying oil (display such as "It's time to change it"). The notification unit 7 can also notify the user of the current deterioration level of the frying oil (display such as "The current deterioration level is 50%") and the remaining usable time estimated from the deterioration level (display such as "20 hours remaining usable").

An example of the notification unit 7 is a speaker, and notification can be made by an auditory method such as a voice guide or an alarm. The notification unit 7 may also make a visual notification by displaying an image, text, color, emitting light, or the like. For example, the notification may be made by displaying an image or text using the display unit 5, or by a light-emitting element such as an LED. Notification by the notification unit 7 is not limited to visual or auditory methods, and may be a combination of these or any method that allows the user to objectively recognize when it is time to change the frying oil, such as vibration.

The comparison and determination unit 13 of the control unit 10 compares the acquired sound data with correlation data corresponding to the type of food cooked with the frying oil to determine the degree of deterioration of the frying oil. The sound generated during cooking with the frying oil contained in the oil tank 22 depends on the type of food being cooked. The optimal time to change the frying oil also differs depending on the type of food being cooked.

The correlation data is pre-stored in the storage unit 6. When making a comparison, the comparison and determination unit 13 acquires the correlation data from the storage unit 6 and determines the degree of deterioration of the frying oil. Note that although the correlation data is created by the machine learning unit 14, it does not necessarily have to be created within the deterioration prediction device 1, and correlation data provided from outside may be used.

When the control unit 10 determines that the deterioration level of the frying oil exceeds a predetermined threshold value according to the type of food, the control unit 10 controls the notification unit 7 to issue a notification. The threshold value is determined in advance for each type of food. The threshold value may be changeable by the user as appropriate. Furthermore, multiple threshold values may be set.

Next, a flowchart of the deterioration determination of frying oil by the deterioration prediction device 1 will be described with reference to Fig. 3. Fig. 3 is a flowchart in the case where a threshold value serving as a guide for replacing frying oil is set in advance.

First, the user obtains information about the food to be cooked and performs the necessary settings (STEP 10). Since the sound generated during cooking differs depending on whether the food being fried is a croquette or tempura, the deterioration prediction device 1 is set according to the type of food being fried. Then, the process proceeds to STEP 20.

In STEP 20, index data is created from the sound during cooking. Specifically, the sound (sound data) during cooking is acquired by the sound data acquisition unit 2 and transmitted to the processing unit 3, and index data such as the frequency average (f_mean) is created. Then, the process proceeds to STEP 30.

In STEP 30, correlation data is obtained from the storage unit. This correlation data is required in the next step of determining the degree of deterioration of the frying oil. Then, the process proceeds to STEP 40.

In STEP 40, the two sets of data are compared to determine the degree of deterioration of the frying oil. Specifically, the comparison and determination unit 13 of the control unit 10 compares the acoustic data with the correlation data. Then, the process proceeds to STEP 50.

Next, it is determined whether the deterioration level of the frying oil has exceeded a predetermined threshold value (STEP 50). This threshold value differs depending on the type of fried food, but if the threshold value has been exceeded, the process proceeds to STEP 60, and if the threshold value has not been exceeded, the process returns to STEP 20.

If the deterioration level of the frying oil exceeds the predetermined threshold (STEP 50: YES), the user is notified of the fact (STEP 60). Specifically, the notification unit 7 notifies the user to prompt the user to replace the frying oil. Then, the series of processes is terminated.

[Second embodiment]
Next, an overview of a deterioration prediction system 100 according to a second embodiment of the present invention will be described with reference to Fig. 4. The deterioration prediction system 100 is mainly composed of a detection device 30 and a machine learning device 40. The detection device 30 and the machine learning device 40 are connected via a network NW and can transmit and receive various data to and from each other.

The detection device 30 includes an acoustic data acquisition unit 2, an input unit 4, a display unit 5, a storage unit 6, a notification unit 7, a communication unit 8, and a control unit 10. The control unit 10 also includes a comparison/determination unit 13. Note that each component except for the communication unit 8 is the same as the component of the processing unit 3 of the first embodiment, and therefore description thereof will be omitted.

In the detection device 30, when the sound data acquisition unit 2 acquires the sound generated when cooking croquettes, tempura, etc., the comparison and judgment unit 13 compares the acquired sound data with correlation data corresponding to the type of food being cooked, and judges the degree of deterioration of the frying oil.

Furthermore, the communication unit 8 (the "first communication unit" of the present invention) automatically transmits the acoustic data to the machine learning device 40 via the network NW. This communication may be wired or wireless, such as Wi-Fi (registered trademark) or Bluetooth (registered trademark). In the deterioration prediction system 100, since only the detection device 30 is required in the store or factory (near the oil tank 22), the device can be made compact.

The machine learning device 40 has a communication unit 48 (a "second communication unit" of the present invention) and a learning model creation unit 50. The acoustic data is automatically received by the communication unit 48 of the machine learning device 40. The machine learning device 40 may be installed in a location separate from the flyer 20. Of course, the detection device 30 and the machine learning device 40 may be an integrated system.

The learning model creation unit 50 has an index extraction unit 51, a storage unit 52, and a calibration curve creation unit 53. The index extraction unit 51 extracts index data related to frying oil deterioration from the received acoustic data, and the index data is stored in the storage unit 52. The calibration curve creation unit 53 performs so-called supervised learning to create a calibration curve (model formula) from the stored index data (explanatory variables) by linear regression analysis.

Types of linear regression (analysis) include simple regression, multiple regression, partial least squares (PLS) regression, and orthogonal partial least squares (OPLS) regression, and one or more types selected from these can be used.

Simple regression is a method to predict one dependent variable with one explanatory variable, while multiple regression is a method to predict one dependent variable with multiple explanatory variables. In addition, (orthogonal projection) partial least squares regression is a method to predict one dependent variable with multiple explanatory variables.
1601259771323_0
In addition, (orthogonal projection) partial least squares regression is a method suitable when the number of explanatory variables is greater than the number of samples and when the correlation between explanatory variables is high.

5A and 5B show the relationship between the calibration curve obtained by machine learning and the heating time (predicted value and actual measured value) of the test data.

The straight line M1 in Fig. 5A is a calibration curve (model formula) obtained by simple regression analysis using the frequency mean (f_mean). In the graph, the horizontal axis is the predicted value of the heating time [h], the vertical axis is the actual value of the heating time [h], and the "○" marks in the figure are plots of the predicted values obtained from the frequency mean (f_mean).

5B shows a list of the current heating time (actual frying time), the average predicted value and the standard deviation of the five times. For example, the average predicted value for the actual heating time of 8 [h] was 8.9 [h], and the standard deviation at this time was 1.4. The predicted values were approximately in the vicinity of the line M1 (see FIG. 5A), and the variation was relatively small, confirming that the calibration curve obtained by the simple regression analysis had a certain degree of accuracy.

6A and 6B show the relationship between the calibration curve obtained by machine learning and the acid values (predicted values and actual measured values) of the test data.

The straight line M2 in Fig. 6A is a calibration curve (model formula) obtained by multiple regression analysis using the frequency mean (f_mean) and the flatness (f_sfm) of the frequency spectrum. In the graph, the horizontal axis is the predicted value of the acid value, and the vertical axis is the actually measured value of the acid value. The "○" marks in the figure are plots of the predicted value of the acid value obtained from the frequency mean (f_mean) and the flatness (f_sfm).

6B shows a list of the heating time, the measured acid value, the average of the five predicted values, and the standard deviation. For example, the measured acid value for the heating time of 8 [h] was 0.16, the average predicted value was 0.11, and the standard deviation was 0.10. Some of the predicted acid value values were on the line M2 (see FIG. 6A), and the variation was small, confirming that the calibration curve obtained by multiple regression analysis was highly accurate.

7A and 7B show the relationship between the calibration curve obtained by machine learning and the heating time (predicted value and actual measured value) of the test data.

Line M3 in Fig. 7A is a calibration curve (model formula) obtained by orthogonal projection partial least squares regression (OPLS) analysis. In the graph, the horizontal axis is the predicted value of heating time [h], and the vertical axis is the actual value of heating time [h]. The "○" marks in the figure are plots of the predicted value of heating time obtained from the frequency mean (f_mean).

7B shows a list of the heating time (actual frying time), the average predicted value and the standard deviation of the five times. For example, the average predicted value for the actual heating time of 8 [h] was 9.0 [h], and the standard deviation was 1.8. Some of the predicted values were on the line M3 (see FIG. 7A), and the variation was relatively small, so a certain degree of accuracy was observed for the calibration curve obtained by the orthogonal projection partial least squares regression analysis.

8A and 8B show the relationship between the calibration curve obtained by machine learning and the acid value (predicted value and actual value) of the test data. Here, the "acid value" is the value measured by the Standard Method for the Analysis of Fats, Oils and Related Materials 2.3.1-2013.

The straight line M4 in Fig. 8A is a calibration curve (model formula) obtained by orthogonal projection partial least squares regression (OPLS) analysis. In the graph, the horizontal axis is the predicted value of the acid value, and the vertical axis is the measured value of the acid value. The "○" marks in the figure are plots of the predicted value of the acid value obtained from index data such as the frequency mean (f_mean).

8B shows a list of the heating time, the measured acid value, the average of the five predicted values, and the standard deviation. For example, the measured acid value for the heating time of 8 [h] was 0.16, the average predicted value was 0.13, and the standard deviation was 0.12. The predicted acid value had a small variation, confirming that the calibration curve obtained by the orthogonal projection partial least squares regression analysis had high accuracy.

9A and 9B show the relationship between the calibration curve obtained by machine learning and the color of the test data (predicted value and actual value). The "color" here refers to the color tone of the frying oil, and indicates "Y+10R" measured by the Standard Method for the Analysis of Fats, Oils and Related Materials 2.2.1.1-1996.

Line M5 in Fig. 9A is a calibration curve (model formula) obtained by partial least squares regression (PLS) analysis. The horizontal axis of the graph is the predicted color value, and the vertical axis is the actual measured color value. The "○" marks in the figure are plots of the predicted color value obtained from index data such as the frequency mean (f_mean).

9B shows a list of the current heating time, the actual color measurement, the average of the five predicted values, and the standard deviation. For example, the actual color measurement for the heating time of 8 [h] was 6.5, the average predicted value was 6.9, and the standard deviation was 1.6. Since the predicted color values showed relatively small variations, it was confirmed that the calibration curve obtained by partial least squares regression analysis had a certain degree of accuracy.

10A and 10B show the relationship between the calibration curve obtained by machine learning and the viscosity increase rate (predicted value and actual value) of the test data. The "viscosity" here is a value indicating the degree of stickiness (viscosity) of frying oil measured with a commercially available viscometer, for example, an E-type viscometer (TVE-25H: manufactured by Toki Sangyo Co., Ltd.), and this time, the viscosity increase rate [%] with respect to the heating time was investigated.

If the measured viscosity of frying oil when it is first used (viscosity at the start of use) is Vs, then as the oil is used to fry food repeatedly, deterioration progresses and the viscosity of the frying oil increases. If the measured viscosity after the oil has been used is Vt, the "viscosity increase rate" is defined as the ratio of the increase in viscosity to Vs (= Vt - Vs).

10A is a calibration curve (model formula) obtained by partial least squares regression (PLS) analysis. The horizontal axis is the predicted value [%] of the viscosity increase rate, and the vertical axis is the actual value [%] of the viscosity increase rate. The "○" marks in the figure are plots of the predicted values of the viscosity increase rate obtained from index data such as the frequency mean (f_mean).

10B shows a list of the current heating time, the measured viscosity increase rate, the average of the five predicted values, and the standard deviation. For example, the measured viscosity increase rate for the measured heating time of 8 [h] was 3.52, the average predicted value was 3.87, and the standard deviation was 0.57. The plotted predicted values of the viscosity increase rate showed little variation, confirming that the calibration curve obtained by partial least squares regression analysis was highly accurate.

As described above, the calibration curve creation unit 53 creates a calibration curve from the index data by linear regression analysis, and any of simple regression, multiple regression, partial least squares (PLS) regression, and orthogonal projection partial least squares (OPLS) regression may be used as the linear regression. The calibration curve actually created showed high accuracy of the deterioration degree even in the results of evaluation of the acid value, color, viscosity increase rate, etc. of frying oil that change with heating time, and it was possible to accurately predict and determine the deterioration of frying oil from the index data related to sound.

In the deterioration prediction system 100 of FIG. 4, the machine learning device 40 may be installed in a remote location away from the store, and the detection device 30 may function as a detection server, and the machine learning device 40 may function as a machine learning server.

In this case, the detection server on the store side is equipped with at least an acoustic data acquisition unit that acquires acoustic data when cooking fried foods, a communication unit that transmits and receives various data (acoustic data, judgment results, etc.) with the machine learning server, and an alarm unit that notifies the degree of deterioration of the frying oil, the timing for replacement, etc. based on the judgment results.

In addition, the remote machine learning server is equipped with at least a communication unit that sends and receives various data to and from the detection server, a learning model creation unit that extracts indicators related to frying oil deterioration from the received acoustic data and performs machine learning using linear regression using the indicators to create a learning model that can determine the deterioration of frying oil, a memory unit that stores the created learning model, and a determination unit that determines the degree of deterioration of the frying oil using the learning model.

According to this configuration, on the machine learning server side, the learning model creation unit performs machine learning using the received acoustic data to create a learning model. Then, the judgment unit judges the degree of deterioration of the frying oil using the learning model and transmits the judgment result to the detection server side. On the detection server side, the notification unit notifies the timing of changing the frying oil based on the judgment result received. In this way, the machine learning server side can share roles such as receiving acoustic data, making a judgment, and returning the judgment result to the detection server.

The learning model is created by the machine learning server and updated, for example, every time new acoustic data is acquired. This eliminates the need to transmit and receive learning models with relatively large data volumes, and allows the store that has the detection server to obtain the timing for changing the frying oil.

[Third embodiment]
Next, an overview of an oil and fat exchange system 200 according to a third embodiment of the present invention will be described with reference to FIG.

11 is a schematic diagram of an oil exchange system 200. As shown in the figure, the oil exchange system 200 is composed of stores A to C equipped with a deterioration prediction device 1 and a fryer 20', a general headquarters H that manages the stores A to C, a manufacturer (oil manufacturer) X of frying oil used in the stores A to C, a distributor (wholesaler or retailer) Y, and a recovery company Z that recovers used oil. Note that since the oil manufacturer may sell directly to customers, the distributor Y is a concept that includes the oil manufacturer.

In the first embodiment, when it was determined that the deterioration level of the frying oil exceeded a predetermined threshold, the notification unit 7 of the deterioration prediction device 1 notified the user of that fact via the speaker, the display unit 5, etc., but in the present embodiment, in addition to such a notification, notification information regarding the deterioration level of the frying oil is output. The notification information may be a content that the deterioration level of the frying oil has exceeded the threshold, or may be a notice that the deterioration level will soon exceed the threshold.

As shown in the figure, when alarm information is notified to headquarters H from store B (an izakaya restaurant), headquarters H analyzes the number of times and frequency of receipt of the alarm information, and, if necessary, makes suggestions or instructs not only store B, but also store A (a tempura restaurant) and store C (a tonkatsu restaurant) on whether frying oil is being used appropriately, whether it is being replaced appropriately, whether there is any waste, etc.

The headquarters H may be in a position to manage not only a plurality of stores but also a plurality of factories in which flyers are installed. The headquarters H may also be located within a store or factory and manage a plurality of flyers within the facility.

This notification information is also sent to frying oil manufacturer X and distributor Y. Manufacturer X receives the notification information and formulates a production plan or sales plan for frying oil. Distributor Y also receives the notification information and places an order for new frying oil, purchasing frying oil P from manufacturer X. Distributor Y then procures new frying oil P for store B (and stores A and C, if necessary).

Furthermore, this notification information is notified to a frying oil recovery company Z (which may be manufacturer X). Upon receiving the notification information, recovery company Z arranges for the recovery of waste oil Q. For example, when recovery company Z receives the notification information a predetermined number of times, recovery company Z visits store B and recovers waste oil Q from oil tank 22 of fryer 20'.

Furthermore, this notification information may be notified to a cleaning worker (not shown). The cleaning worker receives the notification information and visits store B to clean the inside of oil tank 22 of fryer 20' or its vicinity. In this way, in oil exchange system 200, the supply of frying oil to stores A to C, waste oil, and cleaning can be performed quickly.

In addition, if the change of frying oil in the store is automated based on the notification content, the burden on users (store staff) can be further reduced. When notification information is output that the deterioration level of the frying oil has exceeded a threshold, the change of the frying oil starts automatically.

[Fourth embodiment]
Finally, with reference to FIG. 12, an overview of a fryer system 300 according to a fourth embodiment of the present invention will be described.

12 shows a deterioration prediction device 1 and a fryer 20' constituting the fryer system 300 of the present embodiment. Note that, for the fryer 20', the same components as those of the fryer 20 of the first embodiment are denoted by the same reference numerals, and the description thereof will be omitted.

As shown in the figure, a valve control device 61 ("valve control unit" of the present invention) and a new oil tank 62 are installed near the fryer 20'. The new oil tank 62 contains unused frying oil, and the frying oil is supplied to the oil tank 22 via an oil supply pipe 63.

When the valve control device 61 receives notification information from the deterioration prediction device 1 that the frying oil needs to be replaced (because the oil level has reached or exceeded the threshold), it first transmits a control signal to the valve 24' to open the valve 24'. This causes the waste oil to be automatically discharged into the waste oil tank 26 via the oil drain pipe 25.

After a sufficient amount of time has passed, the valve control device 61 again sends a control signal to the valve 24' to close the valve 24'. Thereafter, the valve control device 61 sends a control signal to the valve 64 provided in the middle of the oil supply pipe 63 to open the valve 64. As a result, fresh oil is automatically supplied to the oil tank 22. The amount of fresh oil supplied may be detected by a water level sensor in the new oil tank 62, or the valve 64 may be opened for a predetermined period of time.

According to this embodiment, the valve control device 61 can automatically drain frying oil in use by controlling the valves 24', 64 based on the notification information transmitted from the deterioration prediction device 1. Furthermore, the valve control device 61 controls the automatic supply of new oil from the new oil tank 62, thereby reducing the workload of the user from checking the deterioration degree of the frying oil and discarding it as waste oil until new oil is supplied.

The deterioration prediction device, deterioration prediction system, and oil exchange system described above are merely examples of embodiments of the present invention, and may be modified as appropriate according to the application, purpose, etc. In this example, the frequency mean (f_mean) and the flatness due to the frequency spectrum (f_sfm) are extracted from the acoustic data and a regression analysis is performed, but the frequency standard deviation (f_sd), dominant frequency (dfnum), etc. can also be applied to predict deterioration of frying oil.

Furthermore, the role of each component device in the deterioration prediction system can be changed. In the deterioration prediction system 100 shown in Fig. 4, the detection device 30 and the machine learning device 40 are separated as separate devices, but acoustic data and learning models have large data volumes, and communication takes time and costs. For this reason, a new control device can be provided that can remotely instruct a detection device with a built-in machine learning unit and receive notification information such as the deterioration degree of frying oil from the detection device.

1...deterioration prediction device, 2...acoustic data acquisition unit, 3...processing unit, 4...input unit, 5...display unit, 6...memory unit, 7...alarm unit, 8...communication unit (first communication unit), 10...control unit, 11, 51...index extraction unit, 12...result reception unit, 13...comparison and judgment unit, 14...machine learning unit, 20, 20'...fryer, 21...cabinet, 22...oil tank, 23...heater, 24, 24', 64...valve, 25...oil drain pipe, 26...waste oil tank, 30...detection device, 40...machine learning device, 48...communication unit (second communication unit), 50...learning model creation unit, 52...memory unit, 53...calibration curve creation unit, 61...valve control device, 62...new oil tank, 63...fuel supply pipe, 100...deterioration prediction system, 200...oil change system, 300...fryer system.

Claims (13)

A deterioration prediction device for predicting a deterioration degree of edible oils and fats,
an acoustic data acquisition unit that acquires acoustic data during cooking of fried food using the oil contained in the oil vat;
A storage means stores correlation data indicating a correlation between acoustic data and a deterioration degree of the oil for each type of fried food in association with each other, the correlation data acquisition unit acquiring the correlation data stored in association with the type of fried food being cooked with the oil;
A determination unit that determines a deterioration degree of the oil and grease based on the acoustic data acquired by the acoustic data acquisition unit and the correlation data acquired by the correlation data acquisition unit ;
A deterioration prediction device comprising: Further provided is an alarm unit that notifies the degree of deterioration of the oil or the timing of replacing the oil,
The deterioration prediction device according to claim 1, characterized in that the notification unit issues the notification when the determination unit determines that the degree of deterioration of the oil has exceeded a predetermined replacement threshold based on the degree of deterioration of the oil. The acoustic data used by the determination unit to determine the degree of deterioration of the oil and grease is an index extracted from the acoustic data,
3. The deterioration prediction device according to claim 1, wherein the index is one or more selected from a frequency mean, a frequency standard deviation, a frequency median, a frequency standard error, a frequency mode, a frequency first quartile, a frequency third quartile, a frequency quartile range, a frequency centroid, a frequency skewness, a frequency kurtosis, a frequency spectrum flatness, a frequency spectrum entropy, a frequency spectrum precision, an acoustic complexity index, an acoustic entropy, and a dominant frequency. A deterioration prediction system for predicting a deterioration degree of edible oils and fats, comprising a detection device and a machine learning device,
The detection device includes:
an acoustic data acquisition unit that acquires acoustic data when cooking fried food using the oil contained in the oil vat;
A memory unit that stores a learning model capable of determining deterioration of the oil and fat, the learning model being created by the machine learning device;
A determination unit that determines a degree of deterioration of the oil and grease from the acoustic data using the learning model,
The machine learning device includes:
A learning model creation unit that extracts an index related to the deterioration of the oil and fat from the acoustic data acquired by the acoustic data acquisition unit, performs machine learning by linear regression using the index, and creates the learning model;
A deterioration prediction system comprising: 5. The degradation prediction system according to claim 4, wherein the linear regression is one or more selected from a simple regression, a multiple regression, a partial least squares (PLS) regression, and an orthogonal projection partial least squares (OPLS) regression. The deterioration prediction system according to claim 4 or 5, characterized in that the detection device and the machine learning device are integrated together. The deterioration prediction system according to claim 4 or 5, characterized in that the detection device is installed near the oil tank in a store or factory, and the machine learning device is installed in a remote location away from the store or factory. the detection device includes a first communication unit that transmits the acoustic data acquired by the acoustic data acquisition unit to the machine learning device;
The deterioration prediction system according to claim 4 , 5 or 7 , wherein the machine learning device includes a second communication unit that receives the acoustic data from the detection device. 9. The deterioration prediction system according to claim 8, wherein the first communication unit and the second communication unit are capable of wireless communication. A deterioration prediction method for predicting the deterioration degree of edible oils and fats, comprising:
An acoustic data acquisition step of acquiring acoustic data during cooking of fried food using the oil contained in the oil vat ;
A correlation data acquisition step of acquiring the correlation data stored in a storage means that corresponds to the type of fried food being cooked with the oil, the correlation data being stored in correspondence with the type of fried food being cooked with the oil;
A determination step of determining a deterioration degree of the oil and grease based on the acoustic data acquired in the acoustic data acquisition step and the correlation data acquired in the correlation data acquisition step ;
A deterioration prediction method comprising: An oil and grease exchange system characterized by notifying one or more of an oil and grease distributor, an oil and grease manufacturer, a store headquarters, a factory headquarters, a waste oil collection company, and a cleaning company of the notification information regarding the degree of deterioration of the oil and grease output from the deterioration prediction device described in any one of claims 1 to 3 .
A valve control unit that controls a valve provided in an oil tank based on notification information regarding a deterioration degree of the oil and grease output from the deterioration prediction device according to any one of claims 1 to 3,
The fryer system is characterized in that the valve control unit automatically discards the oil and fat contained in the oil tank. The fryer system according to claim 12, wherein the valve control unit automatically supplies fresh oil to the oil tank.

Images